Public: anyone can read this; the crew can change it.

The role-aware manifest has a byte budget per role, pinned by a test: /llms.txt is 10.3 KB for an anonymous reader #14285

opened this

Check · Review · Land · Live — No review yet. Next: reviewer, no review yet.

Ready to land

/llms.txt now has a byte budget per role, and a new test, test/llms-budget.test.ts, fails when any role goes over. The anonymous text is about 480 bytes smaller (8,978 → 8,500). The new test, the other llms, standards, agent-card, muse-journey, mcp-card and mcp-size tests, and tsc were green on my last full run.

Notes · 2 files · +76 −4

look green — it answered, no tests were run, ready for a look deploy 4178897

Check passeddeploy 4178897

Website spec: 26 findings (report only)

Look

10 pages beside main · 0 new to this pull

/apps

main

at 4adf612 ·

this pull

In the dark

main

at 4adf612 ·

this pull

at 4178897 ·

/docs

main

at 4adf612 ·

this pull

In the dark

main

at 4adf612 ·

this pull

at 4178897 ·

/explore

main

at 4adf612 ·

this pull

In the dark

main

at 4adf612 ·

this pull

at 4178897 ·

/looks/failed

main

at 4adf612 ·

this pull

In the dark

main

at 4adf612 ·

this pull

at 4178897 ·

/looks/missing

main

at 4adf612 ·

this pull

In the dark

main

at 4adf612 ·

this pull

at 4178897 ·

/nope/nothing

main

at 4adf612 ·

this pull

In the dark

main

at 4adf612 ·

this pull

at 4178897 ·

/notes

main

at 4adf612 ·

this pull

In the dark

main

at 4adf612 ·

this pull

at 4178897 ·

/releases

main

at 4adf612 ·

this pull

In the dark

main

at 4adf612 ·

this pull

at 4178897 ·

/review

main

at 4adf612 ·

this pull

In the dark

main

at 4adf612 ·

this pull

at 4178897 ·

/security

main

at 4adf612 ·

this pull

In the dark

main

at 4adf612 ·

this pull

at 4178897 ·

opened this

/llms.txt now has a byte budget per role, and a new test, test/llms-budget.test.ts, fails when any role goes over. The anonymous text is about 480 bytes smaller (8,978 → 8,500). The new test, the other llms, standards, agent-card, muse-journey, mcp-card and mcp-size tests, and tsc were green on my last full run. bun scripts/gate.ts --fix rewrapped comments in one file, and that fix is in the commit.

Sizes on the test's loopback origin (http://localhost:<port>):

RoleBeforeAfterBudget
Anonymous8,9788,5008,500
Signed-in (rw token)79,39679,26579,300
Admin90,82990,69890,700
  • Budget rule: each budget is today's size rounded up to the next hundred, not down, because a round-down budget fails against the size it was measured from. The anonymous text has no headroom: it measures exactly 8,500.
  • Production size: the 10.3 KB production figure is larger than these because production's origin is longer and the origin appears about 45 times. The ceilings are on the loopback origin, not on that figure.
  • Repeated lines: none appear in the anonymous text, and the test checks that.
  • Doors and guide: the test also checks that the MCP, OpenAPI and report-a-problem links are still there, and that each guide page is still linked, one line each.

What I trimmed for the anonymous reader (src/web/standards.ts):

  • I dropped the "Connect an assistant that uses tools" line. It repeated the MCP URL already in the header. This one also affects signed-in readers, which is why their sizes fell by about 130 bytes.
  • The "For agents with an account" section is now one API line for strangers. It points to the MCP card and to the API catalog, which lists the other discovery URLs.
  • The OpenAPI line with the ?tag= list stays, so the two lines test/llms-anonymous.test.ts and test/standards.test.ts check are still there.
  • The full set of discovery lines (server card, agent card, catalog) is unchanged for signed-in readers.

Not done:

  • Signed-in sizes: I didn't shrink them. At 79 KB and 91 KB, the tool list dominates, and cutting it is a separate decision. The new test only pins those sizes where they are.
  • Anonymous text left in place: the Plans table and the "Run it yourself" bullet are still there. Existing tests and the three doors depend on that wording.

Try it: curl -s $FORGE_URL/llms.txt | wc -c, once without a token and once with one, then bun test test/llms-budget.test.ts.

Closes #14272

commented

/llms.txt now has a byte budget per role, and a new test, test/llms-budget.test.ts, fails when any role goes over. The anonymous text is about 480 bytes smaller (8,978 → 8,500). The new test, the other llms, standards, agent-card, muse-journey, mcp-card and mcp-size tests, and tsc were green on my last full run. bun scripts/gate.ts --fix rewrapped comments in one file, and that fix is in the commit.

Sizes on the test's loopback origin (http://localhost:<port>):

RoleBeforeAfterBudget
Anonymous8,9788,5008,500
Signed-in (rw token)79,39679,26579,300
Admin90,82990,69890,700
  • Budget rule: each budget is today's size rounded up to the next hundred, not down, because a round-down budget fails against the size it was measured from. The anonymous text has no headroom: it measures exactly 8,500.
  • Production size: the 10.3 KB production figure is larger than these because production's origin is longer and the origin appears about 45 times. The ceilings are on the loopback origin, not on that figure.
  • Repeated lines: none appear in the anonymous text, and the test checks that.
  • Doors and guide: the test also checks that the MCP, OpenAPI and report-a-problem links are still there, and that each guide page is still linked, one line each.

What I trimmed for the anonymous reader (src/web/standards.ts):

  • I dropped the "Connect an assistant that uses tools" line. It repeated the MCP URL already in the header. This one also affects signed-in readers, which is why their sizes fell by about 130 bytes.
  • The "For agents with an account" section is now one API line for strangers. It points to the MCP card and to the API catalog, which lists the other discovery URLs.
  • The OpenAPI line with the ?tag= list stays, so the two lines test/llms-anonymous.test.ts and test/standards.test.ts check are still there.
  • The full set of discovery lines (server card, agent card, catalog) is unchanged for signed-in readers.

Not done:

  • Signed-in sizes: I didn't shrink them. At 79 KB and 91 KB, the tool list dominates, and cutting it is a separate decision. The new test only pins those sizes where they are.
  • Anonymous text left in place: the Plans table and the "Run it yourself" bullet are still there. Existing tests and the three doors depend on that wording.

Try it: curl -s $FORGE_URL/llms.txt | wc -c, once without a token and once with one, then bun test test/llms-budget.test.ts.

commented

<!-- look-pages 4178897e6351 --> Look pages for 4178897, from the gate:

  • look pages (inferred): this pull changes src/web/standards.ts; the forge pictures / (src/web/standards.ts), /docs (src/web/standards.ts), /explore (src/web/standards.ts) and 25 more, the pages whose routes import it. A "Looks: <path>" or "No look: <why>" line in a commit message overrides
Sign in to join this conversation.

↑↓ navigate ↵ open ⌘↵ its app esc close

Keyboard shortcuts

Docs